Update app.py

app.py

@@ -3,7 +3,7 @@ import base64
 import io
 from huggingface_hub import InferenceClient
 from gtts import gTTS
-from audiorecorder import audiorecorder
+import sounddevice as sd
 import speech_recognition as sr
 from pydub import AudioSegment
 
@@ -11,15 +11,10 @@ if "history" not in st.session_state:
     st.session_state.history = []
 
 recognizer = sr.Recognizer()
-microphone = sr.Microphone()
 
-# reconociendo voz
+# Reconociendo voz con VAD
 def recognize_speech_with_vad(audio_data, show_messages=True):
     try:
-        with sr.AudioFile(audio_data) as source:
-            audio_data = recognizer.record(source, vad_enabled=True)
-            st.success("Fin de la grabación. Procesando audio...")
-
         audio_text = recognizer.recognize_google(audio_data, language="es-ES")
 
         if show_messages:
@@ -35,7 +30,7 @@ def recognize_speech_with_vad(audio_data, show_messages=True):
 
     return audio_text
     
-# preparando entrada para el modelo de lenguaje
+# Preparando entrada para el modelo de lenguaje
 def format_prompt(message, history):
     prompt = "<s>"
 
@@ -46,7 +41,7 @@ def format_prompt(message, history):
     prompt += f"[INST] {message} [/INST]"
     return prompt
 
-# generando respuesta en texto
+# Generando respuesta en texto
 def generate(audio_text, history, temperature=None, max_new_tokens=512, top_p=0.95, repetition_penalty=1.0):
     client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
 
@@ -75,7 +70,7 @@ def generate(audio_text, history, temperature=None, max_new_tokens=512, top_p=0.
     audio_file = text_to_speech(response, speed=1.3)
     return response, audio_file
 
-# texto a voz
+# Texto a voz
 def text_to_speech(text, speed=1.3):
     tts = gTTS(text=text, lang='es')
     audio_fp = io.BytesIO()
@@ -88,28 +83,34 @@ def text_to_speech(text, speed=1.3):
     modified_audio_fp.seek(0)
     return modified_audio_fp
 
-# reproductor de texto a voz
+# Reproductor de texto a voz
 def audio_player_markup(audio_file):
     return f"""
         <audio autoplay="autoplay" controls="controls" src="data:audio/mp3;base64,{base64.b64encode(audio_file.read()).decode()}" type="audio/mp3" id="audio_player"></audio>
     """
 
-# interfaz de usuario
+# Interfaz de usuario
 def main():
     st.title("Chatbot de Voz a Voz")
-    audio_data = audiorecorder("Deteniendo la grabación...", vad_enabled=True)
+    
+    # Configuración de dispositivos de entrada
+    input_devices = sd.query_devices(kind='input')
+    selected_device = st.selectbox("Selecciona tu micrófono:", [device['name'] for device in input_devices])
+
+    # Captura de audio con sounddevice
+    with sd.InputStream(device=input_devices[st.session_state.selected_device]['index'], channels=2, dtype='int16', callback=None):
+        audio_data = sd.rec(int(44100 * 5), samplerate=44100, channels=2, dtype='int16')
+        sd.wait()
 
-    if not audio_data.empty():
-        st.audio(audio_data.export().read(), format="audio/wav")
-        audio_data.export("audio.wav", format="wav")
-        audio_text = recognize_speech_with_vad("audio.wav")
+    st.audio(audio_data, format="audio/wav")
+    audio_text = recognize_speech_with_vad(audio_data)
 
-        if audio_text:
-            st.success("Frase detectada. Procesando audio...")
-            output, audio_file = generate(audio_text, history=st.session_state.history)
+    if audio_text:
+        st.success("Frase detectada. Procesando audio...")
+        output, audio_file = generate(audio_text, history=st.session_state.history)
 
-            if audio_file is not None:
-                st.markdown(audio_player_markup(audio_file), unsafe_allow_html=True)
+        if audio_file is not None:
+            st.markdown(audio_player_markup(audio_file), unsafe_allow_html=True)
 
 if __name__ == "__main__":
     main()